Huawei open-sources TensorFlow competitor MindSpore

By March 31, 2020 No Comments

Huawei has introduced that its framework for AI app building MindSpore is now open supply and to be had on GiHub and Gitee.

The light-weight suite is very similar to Google’s TensorFlow and Fb’s PyTorch because it lowers the barrier to access for builders taking a look so as to add AI to their apps.

Leader scientist at Huawei MindSpore and an IEEE Fellow, Professor Chen Li defined how MindSpore can scale throughout units in a press unlock, pronouncing:

  • Huawei says it’s again to 90 in step with cent capability
  • Google launches device finding out framework for coaching quantum fashions
  • Amazon trialing AI-based buyer fortify brokers

“MindSpore natively adapts to all situations around the software, edge, and cloud. We put in force AI Algorithms As Code thru on-demand collaboration for more straightforward type building, and supply state-of-the-art applied sciences, and co-optimization with Huawei Ascend AI processors to strengthen runtime potency and computing efficiency. We additionally fortify different processors equivalent to GPU and CPU.”


MindSpore already has the backing of quite a lot of companions together with the College of Edinburgh, Peking College, Imperial School London and the robotics startup Milvus.


The framework is in a position to run on processors, graphics playing cards and devoted neural processing gadgets equivalent to the only in Huawei’s personal Ascend AI chips. MindSpore additionally has 20 % fewer strains of codes than different frameworks when coping with herbal language processing fashions and the corporate claims that this results in a median potency spice up of 50 %. Moreover, the framework helps parallel coaching throughout {hardware} in addition to dynamic debugging which allows builders to isolate insects whilst taking much less time to coach AI fashions.

One different fascinating factor about MindSpore is the truth that it does not procedure any knowledge by itself however as a substitute, ingests gradient and type data that has already been processed. Because of this, the framework preserves delicate knowledge even in “cross-scenario” environments whilst making sure that fashions stay tough.

MindSpore lately calls for Python 3.7+ to run however the framework will quickly fortify different languages equivalent to C++, Rust and Julia. 

  • Now we have additionally highlighted the most productive cloud web hosting services and products

By means of VentureBeat